FEGGNN: Feature-Enhanced Gated Graph Neural Network for Robust Few-Shot Skin Disease Classification.

Accurate and timely classification of skin diseases is essential for effective dermatological diagnosis. However, the limited availability of annotated images, particularly for rare or novel conditions, poses a significant challenge. Although few-shot learning (FSL) methods in computer-aided diagnosis (CAD) can decrease the dependence on extensive labeled data, their efficacy is often diminished by these challenges, particularly the catastrophic forgetting defect during the sequence of few-shot tasks. To address these challenges, we propose a Feature Enhanced Gated Graph Neural Network (FEGGNN) framework to improve the few-shot classification of skin diseases. The FEGGNN leverages an efficient Asymmetric Convolutional Network (ACNet) to extract high-quality feature maps from skin lesion images, which are subsequently used to construct a graph where nodes represent feature vectors and edges indicate similarities between samples. The core of FEGGNN consists of multiple aggregation blocks within the Graph Neural Network (GNN) framework, which iteratively refine node and edge features. Each block updates node features by aggregating information from neighboring nodes, weighted by edge features, to capture contextual relationships. Simultaneously, Gated Recurrent Units (GRUs) model long-term dependencies across tasks, enabling effective knowledge transfer and mitigating catastrophic forgetting. The Efficient Channel Attention (ECA) mechanism further enhances edge feature updates by focusing on the most relevant feature channels, optimizing edge weight computation. This iterative refinement process enables FEGGNN to progressively enhance feature representations, ensuring robust performance in diverse few-shot classification tasks. FEGGNN's superior ability to generalize to unseen classes is demonstrated by its state-of-the-art performance, achieving 84.90% accuracy on Derm7pt and 95.19% on SD-198 in 2-way 5-shot settings.